Using Social Media Monitoring Data to Forecast Online Word-of-Mouth Valence: A Network Autoregressive Approach

Managers increasingly use social media for marketing research, particularly to monitor what consumers think about brands. Although social media monitoring can provide rich insights into consumer attitudes, marketers typically use it in a backward-looking manner — that is, to measure past online word-of-mouth (WOM) valence (i.e., sentiment). This article proposes a novel method for using social media monitoring in a forward-looking manner to forecast brands’ future online WOM valence. The approach takes into account information on related brands based on the premise that consumers’ attitudes toward one brand are likely relative to — and therefore associated with — attitudes toward other brands. The method infers associative relations between brands from social media monitoring data by observing which brands are mentioned at the same time in the same social media sources, thus enabling construction of time-varying brand “networks” for representing interdependencies between brands. The authors test six possible methods for capturing brand interdependencies (Jaccard, Dice, anti-Dice, correlation, normalized correlation, and Euclidean distance) and examine the relative performance of each alternative method with a view to identifying the best approach.